A Morphological Analyzer for Gulf Arabic Verbs

Abstract

We present CALIMA GLF , a Gulf Arabic morphological analyzer currently covering over 2,600 verbal lemmas. We describe in detail the process of building the analyzer starting from phonetic dictionary entries to fully inflected orthographic paradigms and associated lexicon and orthographic variants. We evaluate the coverage of CALIMA GLF against Modern Standard Arabic and Egyptian Arabic analyzers on part of a Gulf Arabic novel. CALIMA GLF verb analysis token recall for identifying correct POS tag outperforms both the Modern Standard Arabic and Egyptian Arabic analyzers by over 27.4% and 16.9% absolute, respectively.

The Palatalisation of /ʤ/ into /j/ in Emirati Arabic (EA): A Rule-Governed or Random Alternation?

“…The linguistic situation of Arabic can be described as complex due to the existence of two types of varieties, namely, Modern Standard Arabic (henceforth MSA) and Colloquial Arabic (henceforth CA) (Khalifa et al 2016). MSA is the official language variety in Arab countries; it is the language used in writing, official speeches, sermons, correspondence and the media, whereas CA is the one used in everyday life.…”

“…In addition, some varieties of CA have received more attention than other varieties, e.g. Egyptian Arabic and Jordanian Arabic are among the dialects that have been examined extensively in comparison with other CA dialects (Khalifa et al 2016;Zibin and Altakhaineh 2016). In contrast, Gulf Arabic (henceforth GA) has not received due attention in linguistic research.…”

Arabic Morphological Analysis Techniques

“… MAGEAD [21,45] CALIMA, both of which handle Arabic dialects (MAGEAD entirely manually designed while CALIMA manually verified the annotated data lexicon using several computational techniques). There are three versions of CALIMA: CALIMAEGY [25], CALIMAGLF [26], and CALIMAstar [27]. Respectively, these cover Egyptian Arabic, Gulf Arabic, and all variants of MSA and Arabic dialects.…”

A System for Diacritizing Four Varieties of Arabic

“…Linguistic Data Consortium (LDC) CallHome corpus, containing 160K words worth of transcripts of informal Egyptian Arabic was among the earlier resources used for DA. Employed technologies varied from manually crafted rules (Vergyri and Kirchhoff, 2004), finite state transducer and support vector machine (Habash et al, 2012;Khalifa et al, 2017;Jarrar et al, 2017), Conditional Random Fields (Darwish et al, 2018, and Deep Neural Networks (Abdelali et al, 2018). While there is no standard dataset for evaluation, recent reported performance on Moroccan and Tunisian was a WER of 2.7% and 3.6% respectively (Abdelali et al, 2018).…”